Hi, I'm Jay Singh, Co-Founder of TestMu AI.

Every technology wave has the same shape. Capability jumps first. Trust catches up later. We saw it with cloud, and we are watching it again with AI.

The gap we kept seeing

⚡ AI writes the code in minutes now

🖱️ But someone still has to open the app and confirm it actually works

⏳ The building got fast. The assurance never did

That gap is the defining problem of this era of software, and it is why we built Kane CLI.

How it closes the loop

📥 Ingests your source or PRD

🧩 Designs the use cases and test scenarios

🌐 Runs them in a real Chrome browser, or on Android emulators and iOS simulators

📸 Returns an evidence pack, AC coverage, and a verdict

Not just "did it run". Did it work, and can you prove it. ✅

We believe the next generation of software teams will not be measured by how fast they build. Everyone will build fast. They will be measured by how fast they can trust what they built. 🚀

Would love to hear how your team is dealing with this gap.

Learn more: https://www.testmuai.com/kane-cli-ph/

Hi, I'm Shantanu Wali, Head of AI Products at TestMu AI.

Most tools give you a test runner. We built an assurance loop:

📥 Ingest → your source, your PRD

🧩 Design → use cases, acceptance criteria, test scenarios

▶️ Run → test.md, executed in a real browser or on a mobile device

📸 Evidence → evidence pack for every run
📊 Coverage → how many ACs are covered

✅ Verdict → shipping confidence, backed by proof

Decisions we fought hard for

💻 Local first, running on your own Chrome
💬 Natural language objectives instead of selectors
🔧 Autoheal, so cosmetic UI changes never break a flow
🤖 Agent-native output, so Claude Code or Cursor can call Kane CLI and read the verdict on their own

And starting today, mobile 📱
v0.8.1 makes mobile a first-class platform. Same setup, same sessions, same evidence model, now on Android emulators and iOS simulators. Run devices list, pick a target, and the AI reasons about mobile screens the way it reasons about web pages.

For the skeptics
Hacker Mode exports any flow to native
Playwright code. Never locked in. 🔓

Ask me anything technical, I will be here all day.

Learn more: testmuai.com/kane-cli-ph

Hi everyone, I'm one of the engineers building Kane CLI. Good to finally have this out in the world.

The part I'm personally most proud of: we built it agent-native from day one. I've spent 2+ years shipping frontend at TestMu AI, and if that taught me anything, it's that writing UI was never the slow part, proving it still works after every change is. AI has made the writing nearly instant, which makes verification the real bottleneck. So we designed for it: every command can emit machine readable output, meaning it's not just humans in a terminal, your coding agent (Claude Code, Cursor, whatever you use) can invoke kane-cli, stream the run, read the verdict, and act on it inside its own loop. AI writes the code, Kane CLI verifies it in a real browser (or on mobile), and the agent gets machine-readable proof back. Closing that loop is what makes autonomous dev workflows actually trustworthy.

The other thing we obsessed over: a pass has to actually mean something. Every verdict ships with an evidence pack; screenshots, console logs, network responses; so a green check is never "the model felt good about it." You can see exactly what was verified, and when something fails you're not guessing.

It's local-first and free to start. Happy to answer any questions in the comments.

Big day for us at TestMu AI

I have watched Kane CLI come together from the inside, and the part that still gets me is watching it in action. You type what you want verified in natural language, a real Chrome window opens, it works through the flow, and comes back with an evidence pack and a verdict. No selectors written. No framework set up.

And the timing of this launch could not be better. As of today it does the same thing on Android emulators and iOS simulators. Same commands, same evidence, now on mobile.

The team has poured months into getting the details right, determinism, Autoheal, agent-native output, and it shows.

Proud of this one. If you ship with AI agents, or you are just tired of clicking through your app before every release, give it a run. First verdict lands in minutes.

Learn more: https://www.testmuai.com/kane-cli-ph/

How much context can Kane understand from a PRD before generating the actual browser scenarios?

Hi, I'm Mudit Singh, Co-Founder of TestMu AI.

The thing I love most about Kane CLI: the whole loop works for everyone.

👨‍💻 Developers feed it a PRD and get scenarios, runs, and a verdict before the PR
✍️ Business teams can describe flows in natural language and automate the boring browser stuff
🤖 AI agents run it to verify their own work and read the verdict straight from the output
📱 Mobile teams, starting today: point the same commands at Android emulators and iOS simulators

Every run leaves an evidence pack. Every verdict is backed by AC coverage. Nothing rests on "trust me, I checked".

Getting started is the easy part

The Starter plan is free. One install command, npm, brew, or curl, and your first verdict lands in minutes.

Try it, break it, tell us what is missing. Your feedback this week directly shapes the roadmap.

Learn more: testmuai.com/kane-cli-ph

Kane CLI is the future of agentic verification systems.

how is this different from Claude in Chrome?

Hi, I'm Mayank Bhola, Co-Founder and Head of Products @TestMu AI

It started with a pattern that would not go away. Agent ships the code. PR merges. Tests pass. Three days later someone opens the app and the button does not work.

Everyone kept asking how to make agents write better code. Wrong question. Agents could already open browsers and check their work. The output was the problem. Different result every run. No verdict you could trust. Loop it enough times to be sure and you have burned real money for a maybe.

So we set one non-negotiable bar: determinism

Same flow, same result, every time. A pass is granted only when the expected state is verified through explicit evidence:

• DOM state and URL changes

• Network responses

• Screenshots and console logs

• AC coverage behind every verdict
  

And as of today, that bar holds on mobile too

v0.8.1 brings the same verification contract to Android emulators and iOS simulators. Same evidence, same replay, same verdict you can defend.

This is the layer I wished existed every time a merged PR broke in production. Now it does, on web and mobile.

Break it. Question it. I am here all day.

Learn more: testmuai.com/kane-cli-ph

Testing across browsers and real devices already creates plenty of complexity. Adding agentic reasoning could make that easier to manage.

This is super useful, especially with how much code is being written by AI now.

What has been the most surprising use case you've seen from early users?

Congrats @jay_0687 & team!

Launch day energy is real today 🎉

I have told the Kane story on stages and in campaigns for months, and the reaction is the same every time. People assume they will need to write code or learn a framework. Then they watch someone type a sentence in natural language, a real browser opens, and it comes back with an evidence pack and a verdict.

That moment when the room goes "wait, that's it?" never gets old.

And today the story got a new chapter: the same commands now run on Android emulators and iOS simulators too.

Even outside engineering, I use Kane CLI for my own repetitive browser work. That is the part I love most, it is not just for engineers.
The Starter plan is free, so go break it and tell us what you think. The team is here in the comments all day.

Learn more: https://www.testmuai.com/kane-cli-ph/

I had been looking for such a product that can help me in validation during my development sessions. Kane CLI seems like the perfect tool that will really help me keep my work bug-proof. Amazing!!

Agentic mobile test was the missing piece of the puzzle. Kane CLI seems to be solving that perfectly well!

Could developers trigger Kane directly from CI/CD and block deployments automatically when critical browser workflows fail?

The build-fast, trust-slow gap is very real. I like how Kane CLI focuses not just on running tests, but on producing evidence teams can actually trust and share.

How do you see Kane CLI fitting into an AI coding workflow,should testing happen automatically after every code change?

I really love KaneAI’s conversational approach to generating critical test scenarios from the provided documentation. It frees up my local machine’s resources while helping me explore and work with different aspects of the platform more efficiently. Overall, KaneAI has made the testing process smoother, faster, and more collaborative.

Congrats, team! How does Kane decide whether a natural-language test has actually passed when the page behavior is slightly different than expected?

the ask-tool pause on OTP/CAPTCHA instead of guessing is the right call. since the exported case replays deterministically, what happens when the site's UI shifts a little, a moved button or new copy, does it re-derive from the original NL description or does someone have to re-record?

Does Kane handle logged-in flows automatically, or do we need to set up auth ourselves?

Awesome, i have been needing to formalise our test releases.. this is just what i was looking for.

Does the product integrate with other systems?

the machine-readable verdict for agent loops is the right instinct, that's the same gap I hit running voice model evals: a pass/fail with no provenance just means trusting the model again. does the evidence pack capture intermediate state, or just the final screenshot, when a flaky animation causes a false fail?

AI coding agents desperately need something like this. Writing code is becoming easy; reliably verifying the result is becoming the bottleneck.

Hey Product Hunt. I'm Evan, cofounder of Ito.

Coding agents produce more code than you can review. Existing code review tools only stare at the code, so you still have to review tons of code or build complex internal systems to verify it.

Ito is a runtime code review tool. On every pull request:

• Creates test plan from code, PR description, ticket, your input

• Builds your app in a single-use isolated container with the ability to use credentials, seed data to reach the state that matters, and use external services just as an engineer would on their local machine. Mocks data/code when it doesn’t have access.

• Drives the app with a swarm of agents (click, navigate, API call, db verification)

• Catches runtime issues: concurrency, failure handling, data migration, authentication, etc.

• Posts results on PR with evidence: video, logs, screenshots

We built Ito with a team of 8 MIT engineers, and spent many cycles to arbitrarily build any application. 

Ito gets better the more you use it:

• Responds to feedback

• Faster runtimes every run as Ito learns how to navigate your application and builds tools to manipulate your environment

• Reviews focused on the areas you care about (sees what gets fixed vs ignored)

• Test instruction per org, per repo, or per author

First 100 reviews free, no card. $40/mo after that. Free forever for open source (already supporting >100k stars): app.ito.ai?ph

You can also try out open source PRs without signing up here: https://www.ito.ai/review?ph

Love the runtime angle. How do you handle PRs that touch external services — mock them in the ephemeral env, or spin up real instances?

Hi, I'm Barron Caster, Co-Founder of Ito AI.

The thing I love most about Ito: gives me confidence to ship fast as a non-engineer. I've fully taken over our marketing site and don't have to bother engineers to check my code.

• Vibecoding our entire marketing site & infra (design, content, performance improvements, SEO, etc)

• Knows entire context of our codebase so nothing gets left behind. Can request specific tests if I want them (e.g. focus on mobile usability of the blog)

• Videos of everything working for every change (and feature demo videos coming soon)

Now the whole team is shipping faster.

Easy to try and see how Ito can help your team: app.ito.ai?ph

Congrats to you and the team, Evan! This is a huge problem, and I know it will remain one. Excited for you all to help eng teams solve it.

It's a neat approach that is differentiated from and complimentary to the standard AI review tools. We use it and it has surfaced bugs other tools missed on multiple occasions.

The team moves fast too and is very responsive to feedback. They are a pleasure to work with.

Hi, I'm Geoff, a software engineer on the Ito team. 👋

The thing I love most about Ito: it actually runs the code. 🚀 Every PR I open comes back with a test plan that was

executed, not just read.

⚡ Automates the manual testing I used to do by hand — click through the flow, check the edge cases, confirm nothing else broke. That's hours back every week.

🐛 Catches bugs I would have missed. Real runtime failures that linters and static analysis structurally can't see,

because they never execute anything.

🤖 Makes agentic loops genuinely feasible. When an agent writes the code, something has to verify it actually works

Ito is that step, so I can let the loop run longer and trust what comes out the other end.

Getting to build this as an engineer who also uses it every day has been the most fun I've had on a product. 🎉

Easy to try and see how Ito can help your team: app.ito.ai

Question: How do you compare with Greptiles Trex feature?

Really interesting approach to code review. The fact that Ito actually runs the app and returns runtime evidence makes this feel much more actionable than reviews based purely on diffs. Especially compelling for the edge cases static analysis can miss. Congrats on the launch! 🚀

The runtime approach to code review is really interesting. Instead of just analyzing what the code looks like, Ito actually checks how the application behaves in real scenarios and backs the findings with evidence. The combination of agent swarms and isolated environments makes this especially relevant as AI-generated code keeps increasing.

the "mocks data when it doesn't have access" line is the scary part to me, a mocked call that quietly diverges from the real service gives a false pass that looks more trustworthy than static analysis. do you flag which parts of the evidence pack came from a mock vs a live call?

I like the ephemeral env per PR idea - less guessing from a diff

congrats!

Running the app before reading the diff is the actual difference here, and it creates a question I have not seen a runtime reviewer answer well.

What does Ito post on the PR when the container itself never came up? Build failed, seed data unavailable, external service refused the credential. If that lands as "no runtime issues found", then a broken environment and a clean PR are the same result, and it fails silently in the direction that looks fine.

So: what does a run that could not run report as?

Grayson here, one of Ito's cofounders.

I love that the hard part of Ito is often overlooked: generalizing the ability to run code is the difficult problem. Everyone assumes it's a "deciding what to test" problem (which is definitely hard too!). But the challenge of standing up an actual application is insanely hard. Get that wrong and the smartest agent in the world clicks around a broken login page.

Internal teams at DoorDash or Ramp have spent quarters of engineering time trying to build this internally because they know it's important. Most teams cannot, and shouldn't have to.

That is why runtime review didn't exist until now. Nobody wanted to build the unglamorous layer, and you cannot fake it by bolting a sandbox onto a static reviewer.

That layer is where most of our engineering was spent prior to launch, and it's the reason the reviews surface verifiable bugs instead of suspicions. Just wait to see what else we do :)

Can I run it on my local machine?

Hey Product Hunt 👋 Yuanlin here, founder of Zeabur and now building Nuphos.

Nuphos started from something we learned while building Zeabur.

We used to describe where Zeabur was going as “Your AI DevOps Engineer.”

But eventually we realized:

A PaaS asks you to adapt to its rules. A DevOps engineer adapts to your team’s infrastructure.

They learn the systems you already have, work within your permissions, and follow the way your team operates.

AI DevOps engineers should too.

Nuphos is an AI-native DevOps workspace that brings AI agents into the infrastructure your team already runs, without giving up control.

Our team of fewer than 10 engineers operates across 8+ clouds, 20+ Kubernetes clusters, and more than 10,000 hosts. We’d already been using coding and terminal agents in production, and they were surprisingly capable.

The hard part wasn’t getting an agent to run kubectl or inspect logs. It was everything around the action:

• Whose permissions is it using?

• What can it change?

• Who approved it?

• What exactly changed?

That’s what we built Nuphos around.

With Nuphos, you can connect your existing AWS, GCP, Kubernetes, and observability stack, keep agents read-only by default, require approval for write actions, and keep the context and audit trail shared with your team.

You can try it with your own infrastructure, or use the live demo on our homepage without signing up.

🎁 Use code PH100OFF to get your first month of Nuphos free.

👉 Got questions or want to follow what we’re building? Join our Discord: https://nuphos.ai/dc

And I’d love to hear from people running production:

What kind of experience would make an AI feel like a real DevOps engineer on your team, rather than just another AI agent?

The point that running kubectl or inspecting logs was not the hard part really stood out to me. We already know agents can execute commands. The real challenge is whether a team can trust them in production. Permissions, approvals, and a clear record of what changed are what turn a capable agent into something an engineering team can actually work with. Congrats on launching!

Love the idea of an AI agent that adapts to the team’s existing infrastructure. Congrats to the Nuphos team!

I worked on the approval gate feature. There were a lot of internal debates around how much context to show, when to ask for approval, and what happens if someone interrupts halfway through. Would love feedback from people who deal with production systems every day.

Really interesting approach. How does Nuphos learn the context of a team’s infrastructure over time? For example, can it pick up internal runbooks, past incidents, and team-specific operational conventions?

I’m one of the platform engineers building Nuphos, what concerns me the most is how each infra team adopt our product. I'd love to see the most brutal, realistic way on how every other SREs would use Nuphos. 🙂

Interrupt the agent mid-run, change direction, reject its plan, don't go easy on us. This is exactly how we dogfood and use Nuphos on our infra from day to day. We'd appreciate feedback, apart from just a simple "looks good".

Coming from the marketing side of the team, I’ve been preparing for this launch for a long time.
I’ve watched Nuphos take shape through every piece of copy, every image, and every video we created.

The living runbook is probably the feature I’m most excited about. Too much valuable incident knowledge gets buried in Slack, only to be rediscovered at 2 a.m. months later.

So happy to finally see Nuphos live 🎉
Give it a try! We’d genuinely love to hear any feedback, especially what feels wrong or where it breaks.

agent devops is the way!!! i hate reading raw events

This looks really exciting! One thing I’m curious about: how would you see Nuphos fitting into the workflow of a solo developer or a one-person team?

I build and experiment with multiple small products and AI projects, but I don’t necessarily have a large SaaS infrastructure or a dedicated DevOps team.

Would Nuphos still be useful at this stage? If so, what would be a good real-world use case for someone like me to start with?

Or is Nuphos currently more suited to larger teams that already have a fairly complex production infrastructure?

Would love to understand where a solo builder fits into the Nuphos vision.

We’ve been using @Nuphosas their design partner. This is pretty much how I want AI to work with our production infra. Congrats@yuaanlin

Congrats on the launch!

Really interesting to see agents operating directly against customer infrastructure.

How do you evaluate whether an agent will make the right infrastructure decision across scenarios it hasn’t encountered before, especially failures that would be expensive or risky to reproduce in a real cloud environment?

The point about running kubectl not being the hard part really resonated with me. Agents can already execute commands. The bigger challenge is making sure teams understand what they are doing and feel comfortable giving them access to production. Nuphos seems to be solving the right problem.

Nuphos team here! 👋

My favorite part is the exact moment when monitoring FIREs, Nuphos agent starts investigating, and the context lands in our Slack channel. The agent gets to work automatically, with prod access under-watched, human approved. Being platform engineer myself, the collaborative workspace is what empowers me the most.

Adding a PM's angle to what the team's shared above: the part I keep coming back to internally is the shared context layer. So much DevOps knowledge normally lives in one person's head or one terminal session — when an incident hits, you're paying for that every time. What made Nuphos click for me was watching the agent's understanding of our infra become something the whole team could see and build on, not siloed. That shift from "individual knowledge" to "team context" is subtle but it's honestly the part I'm most excited to keep developing.

Quesstion: Who is the ideal cstomer here? Who do you solve this for? With which customer does this fit the best?

the read-only-by-default plus approval-for-writes model makes sense for trust. during an active incident when minutes matter, does that approval step ever become the bottleneck itself, or is there a break-glass path for that moment?

This feels like the right direction for AI in DevOps. Running commands is the easy part — understanding the infrastructure, permissions, past incidents, and how a team actually operates is much harder.

Really like the read-only-by-default + approval approach too. Agents touching prod need context and guardrails, not just more autonomy. Congrats on the launch!

This looks promising. I would be curious to see how Nuphos handles situations where the agent is unsure or finds conflicting signals. Knowing when to pause and ask a human could be just as important as knowing what action to take.

Part of the team here! One thing I really like about the permission admin design is how it makes the whole privilege escalation flow feel much safer. Instead of access changes being handled loosely or buried in chat, there’s a clearer approval path before anyone gets elevated permissions. That became a big part of how we thought about making admin workflows more trustworthy.

Team member here. My favorite part is actually not the AI—it’s finally having logs, cluster state, deployments, and the rest of the incident context in one place. The agent becomes a lot more useful once everyone is looking at the same thing.

This is a strong step toward practical AI-native DevOps. Giving agents real infrastructure context—while keeping humans in control of investigation and production actions—could save engineering teams a huge amount of time. Excited to see where Nuphos goes!

Been using early builds internally for a while, so today feels pretty special. It still makes mistakes (of course), but having it collect the clues before I jump into an incident has already saved me a lot of tab-hopping.

Launch day! I worked on growth side, it’s fun watching Nuphos evolve. From “can we stop hogging across 10 tools for every incident?”, to something we now use in our own infrastructure, dogfooding every single day. Can't wait to see more devops engineer start using Nuphos, bringing more thoughts on the idea of AIOps!

Hello! 😁

I'm Per, CEO and co-founder of Scrimba, a learning platform for 2+ million devs which I've been running for 10 years.

"Explain" is our attempt at creating the perfect AI tutor. The core idea is:

1. As fast as a chatbot.

2. As engaging as a video.

You type a question and within 2-3 seconds to get a video-like response, with animations, diagrams, images, code, and more. Narrated by a pleasant voice.

It's currently free for everyone during the beta phase 🤑 Our goal is to keep costs as low as humanly possible so that we can continue to offer a generous free plan for students.

Our users have already created over 20,000 explainers during the beta phase. It resonates really well with the type of people who prefer YouTube tutorials over technical articles.

You can also bring "Explain" to the tools you use day-to-day

• ChatGPT Plugin

• MCP support for coding agents like Codex, Claude, Cursor etc

• Chrome Extension

• GitHub: Video Explainers of your Pull Requests

The latter has been a godsend for us internally as we're drowning in PR's these days 😂 Getting a quick video makes it a lot easier for our team to review each others' code.

Looking forward to hear what you think! And please shout out if you have a feature request. I promise we'll ship it within a week!

Here's a few examples:

• How a Circle Becomes Music: https://scrimba.com/explain/guide0csjdqa3n

• Meta explainer about Explain: https://scrimba.com/explain/guide0ien1j3b1

• Understanding CORS (playlist): https://scrimba.com/explain/guide0n4r04ilh

Just tried Explain by asking "how the frontend sends data to the backend" and I was genuinely impressed by how quickly it turned the answer into a clear short video with diagrams and visuals!

What I really liked was that it asked me a question at the end to check whether I actually understood the explanation 🤩 as someone who creates programming education content, I really like that approach! It makes it feel much more like an AI tutor than simply another way of presenting an AI-generated answer.

Really interesting direction for learning technical concepts. Excited to see where you take this - well done, Scrimba team! Keep it up! 💪💪

Oh my goodness, I am blown away.

I made it explain to me MedusaJS.

Wow!!! I had to let my wife try it, she loves it.

I've been giving this a spin! Lot's of fun, and the implementation is fast, intuitive and delightful. I've used it on my code base to understand it better, and a few things I've been curious about.

I have personally tried scrimba explain feature number of times til lnow and it is really amazing. It not only solve your confusion but also force you to think about other aspects as well. Great work by Per and Scrimba team by adding this feature to Scrimba. Really love it. Thanks for sharing with it learners community.

Scrimba explain is one of my favourite scrimba features. It's really helped me grasp concepts I've been struggling with since forever. For example backend development terms like json, API architecture and so on

That is super cool! TIL you can ask it questions in general, not just about arcane JS things!

I just tried it out. Great that it make a video on the subject. Quite a surprise !

Congrats on the launch! For developers / product people interested in using Human Behavior, where can they sign up?

Such a great product. Just curious, how has your own product helped you prioritize features?

Huge congrats on the launch! I really like the problem you're tackling here. There’s so much user data sitting inside products, but actually understanding what users are struggling with is still surprisingly hard. Turning those messy session replays into clear, actionable insights feels genuinely useful. Wishing the whole team a strong launch today — excited to see how Human Behavior evolves!

closing the loop straight to Linear and Slack without forcing another dashboard on us is huge. congrats on the launch!

Thought of building this for myself, glad to see someone else do it! Sent you a DM :)

the dogfood stat is the most convincing part, 16 PRs out of 15k sessions with none arriving as a bug report first. of those 16, how many merged clean vs needed a human to say actually that's expected behavior?

The era of dashboards is coming to an end, great use of AI here. Congrats on the launch!

Really interesting approach to turning product analytics into action rather than another dashboard to monitor. As a founder, I’m curious: how do you see teams balancing autonomous AI actions with the need for human approval when those actions affect real customers?

Looks amazing & congrats on the launch!

the part id want to know about is the emailing. a PR or a linear ticket is reversible, an email to a customer isnt. is there a human in front of that one, or does it clear the same bar as opening a PR?

Congrats on the launch! It resonates. Seeing the drop-off was never the hard part. Understanding why it drops and fixing the issue is the real challenge. Question: can I set different leashes per action? (auto-open PRs and Linear tickets, but keep customer emails behind my approval)

Hey Product Hunt! 👋

I'm Stefano, part of the team building Oasis. We're excited to finally share what we've been working on.

Agents are getting really good. But the way we work with them still feels fundamentally broken.

Today, every agent lives somewhere different. One is in a browser tab. Another is running locally. Another is inside your IDE. We spend our days copying context between tabs, terminals, and chats.

That fragmentation is frustrating on its own. It gets way worse when you try work with your team!

Your teammates can't see the same context. They can't jump into the conversation. They can't reuse your agents. Every person builds their own isolated workflows, and everyone starts from scratch. The more agents you adopt, the more fragmented your AI stack becomes.

Agents shouldn't feel like isolated tools. They should feel like teammates, living alongside you and your team.

That's why we are building Oasis, the shared workspace for humans and agents.

What Oasis does:
- Brings every agent into one place
When you join Oasis, you instantly have access to the best agents in the ecosystem. We partner with leading agent builders so you can deploy Claude Code, Devin, OpenClaw, HeyGen, Higgsfield, or any MCP-compatible agent in a single click.

- Work together in Rooms
In the real world, work happens in meeting rooms. Oasis works the same way.
A Room is a group chat for humans and agents. Instead of scattered chats and isolated workflows, everyone works together from the same context.

- Build a shared memory base
Every conversation, decision, and artifact becomes part of your organization's memory.
New agents don't start from scratch-they plug into the knowledge your team has already built, with permissions enforced from day one.

- Invite your team
AI shouldn't be single-player. Bring your teammates into the same workspace, share agents across your organization, and collaborate with both humans and AI without forwarding links, copying prompts, or rebuilding context.

Thanks for checking us out. We'll be here all day answering questions and would love your feedback.

Join our product hunt slack community to talk to the team building Oasis directly and get first-hand updates: https://joinoasis.com/oasis-comm...

See you in the Oasis 🫡

Really cool product ❤️

Oasis tackles one of the biggest problems with using multiple agents today, everyone ends up working with different context in different places! How do you think about permissions and memory once humans and many different agents are all working inside the same Room? Can teams control exactly what each agent is allowed to remember and reuse? :)

Multiplayer is an interesting concept :) I wonder what the ratio of humans to AI ends up being.

How does one use the coupon code for the credits?

Congrats on the launch :)) humalike.ai would fit amazingly

Love your product idea! Running agents across separate tools means every one of them starts from zero, so nothing you learn in one place ever compounds.
Is there a possibility to scope an agent to specific channels / projects rather than everything the workspace knows?

Hey Stefano! Love the idea of a place where Humans and Agents can work together. Feel that's the way and I'm sure you'll rock it. Wish you all the best!

🔥

Collaborating on .md files is a pain - we paste them into Notion, or open them locally and screenshare. Then after we edit and comment on them, there's no easy way to get the feedback back into the coding agents. So...I built Skilldocs.

I picked the best features of some of my favorite products: @HackMD 's mono viewer, @Figma 's live cursors + follow (no screenshare necessary), Google Docs' real-time highlighting and comments, and @Bear 's beautiful wysiwyg markdown editor. After working on the doc together, leaving comments, and editing it, you can easily copy the diff and comments and send them to your coding agent of choice.

I've started using it for collaborative docs in general. It's really fun having a meeting in it.

Minor features I love:
-Click on a teammates avatar to follow them through a doc or even through switching docs. It's like following in Figma - it feels like screenshare but snappier.

-You can cmd+click to create multiple carets and edit several lines at once (after @Sublime ).

Right now you can import skills via dropping a .md file, pasting markdown, importing from Github, or from your local machine via an MCP. Export is simple: copy the .md of the current doc, diff, or comments. I haven't built proper team source control yet but I've gotten a few requests - I'd love requests and suggestions here.

h/t to @mikekerzhner for early ideation, and @bernatfortet @teamcoscreen @heshamghandour for early testing!

on the source control question - since everyone's already live in the same doc like Figma, I'm not sure you actually want git branches, branching is for when people edit in isolation and need to reconcile later, which is the opposite of what you've built. what I'd actually want is lighter: the ability to name/tag a version before I hand the diff to my coding agent, so if the agent runs with a bad version of the skill I can point back to "the one from tuesday's meeting" instead of reconstructing it from memory. real branching only starts to matter once two agents are independently proposing edits to the same skill without a live human session tying them together, which sounds like a later problem, not a v1 one.

This is awesome! I have the wrong mental model for the MCP server install, though, and could use some help. I clicked on the "Claude Code can work in here" and am following the instructions. To install the MCP server using this command, do I need to check out a local copy of skilldocs source code? (Very sorry if this is obvious in docs that I should have seen.)

```
claude mcp add skilldocs -e SKILLDOCS_TOKEN=<paste-token> -- node <path-to-skilldocs-repo>/mcp/index.js
```

Skilldocs solves a surprisingly awkward collaboration gap! How are you thinking about source control once multiple people and agents are editing the same skill? Will comments and diffs eventually map cleanly back to Git branches/PRs?

That looks amazing. Will check it out. Congratulations to the launch!

Great journey so far @jbrussee, congrats on launching and can't wait to see what else you're going to ship!

Awesome journey @jbrussee let's gooo!! 🚀

upvoting because your tagline is top tier 😂

I’ve been sitting on these illustrations for way too long.

A few years ago, we built an ambitious 2D world and game builder. I spent about a year leading the illustration work for it, creating kits full of tiny buildings, trees, props, characters, rocks, weapons, furniture and all sorts of other bits you could combine into your own worlds.

I loved the product and really believed in it, but it never got the chance to become what we hoped it could be. Leaving thousands of illustrations sitting in old Figma files felt like such a waste.

So, as a little side project, I started giving them a second life.

That became Kitbitz

There are now 2,000+ hand-drawn assets across 13 themed kits, all released under CC0. You can download individual SVGs and PNGs, or grab the full Figma Community libraries with reusable components, color variables and flexible assets built to resize, stretch and tweak without everything going wonky.

And I’m having way too much fun taking it further. I’m already playing with a Figma plugin, MCP support and a terrain/world generator to make the kits even more useful and fun to build with.

Take whatever you want. Make something fun with it. 🚀

This is so cool!! 🥹 So much work and love clearly went into these. Very excited to see what people make with Kitbitz. Congrats @sjoerd_huisman on finally getting it out into the world!! 🚀

I've had the fortune to work with Sjoerd on many projects over several years and got to watch a lot of these get made. Incredible illustrator/designer/dev. What still gets me is how well the pieces fit together... you can build almost any world with them and nothing looks out of place. That part is really hard to pull off

So happy these are finally out. Congrats Sjoerd!!!

With the whole maximalist design trend, this is a great way to level up designs. We're out of the simplistic, sleek era for sure.

Must have for the game-like products!

These could be pretty cool for creating a video game! :D

Hey Product Hunt! 👋

Ever since we started Mem, we've been obsessed with the idea that your mind is a gold mine—full of thoughts, aspirations, and to-dos that never see the light of day again. Today, we're excited to launch Mem Agent, the AI that refuses to let you drop the ball on anything that you care about, grounded in what you tell it and the notes and bits of info you feed it. One unexpected surprise from early users: one thing they have dropped is… their task management apps.

How it works

1. Dump what you need to remember in Mem – meeting notes, brain dumps, files. Text it, Slack it, forward an email, or hit one key to Push-to-Remember with voice.

2. Trust – Mem Agent takes what you've given it and folds it into an ever-evolving model of your current work and priorities, and when it'll be relevant later.

3. Get nudged at the right time – Forget setting due dates or alarms on your phone. Know that you'll get reminded of the thing when the time is right. For the important stuff, Mem won't drop it until you say the word.

How it's different

• Opinionated and works right out of the box: Most AI tools require you to prompt, configure, and piece together skills to be good at following up. We wrote it a charter: the behaviors it’s not allowed to ever give up, so it works on day one.

• It's got receipts: At any point, pop open the Agent Overlay and look into Mem Agent's brain—see what tasks, projects, and goals it's tracking for you. Then close it and get back to not worrying whether someone's got it. (Someone does.)

• Connected apps provide more context, not more noise: When you connect apps (like Gmail), Mem Agent uses them as context to skip reminders for things you've already handled, giving you fewer obsolete reminders to sift through.

• Backed by a shared workspace: Your Mem workspace is where the notes you and Mem create live. Capture new notes, import existing ones, or let Mem take all the notes. It's your shared brain, all cleaned up and organized—no “procrastivity” work of managing a sidebar of folders here.

• Fastest capture and recall: Never lose another thought; summon details deep in your notes with a single keypress using Push to Remember.

• Persistent as hell: It won't let an important thread go cold—not on its watch—unless you tell it to. It'll change tactics or find another time, but it won’t quietly give up.

How to start

• Download the app. (Try Mem Agent with Proactive plan free for 7 days, and don’t forget to use code DROPTHEBALL to get 50% off your first month when signing up on web/desktop.)

• Immediately try Push-to-Remember and just brain dump with your voice—getting every little task off your mind and into Mem.

That’s it. Mem takes it from there—expect a ping at just the right time (“hey, still grabbing pasta for tonight’s dinner?”).

Early feedback from customers

• "I have stopped keeping a visible task list. I tell the agent roughly when and it picks the timing. I stress less about tasks now and this is not what I would have guessed before starting using agent."

• "Please take all my money. This thing is amazing!!! It's like OpenClaw and Mem had a baby!"

• "Agent so far is the best thing I didn't know I needed. My productivity is on another level"

Give it a spin and tell us what you love and what you want to see next—and tell me: what's the one thing you'd want an agent to refuse to let you drop? 👇

Kevin
CEO and Founder of Mem

PS -- I've also shared a longer form comparison of Claude vs. OpenClaw vs. Mem Agent. Check it out: https://x.com/kevinfmoody/status/2087718606281015615

Try it, it might change the way you work

I've been using Mem Agent for several weeks now, and I strongly recommend giving it a try because it brings together - for me at least - a calmer, more personal way of working, reliable project follow-through, a practical bridge between handwritten and digital planning, meaningful health-data context, and a useful cross-checking layer for sources Mem cannot (yet?) access directly:

• It has helped me turn GTD into something I can actually live inside, rather than another planning system. I was able to teach Mem Agent my own way of working through explicit instructions, personal skills, and routines — for example, how to keep the background bookkeeping quiet, bring unclear points to me one at a time, and combine project context, calendar shape, energy/readiness, and real external deadlines into one suggested needle-mover or several suggested tasks fitting the context and time when I ask for it. The structure is maintained underneath, but the next action is chosen intuitively by me in the moment based on what feels doable and whats presented by Mem, which makes my GTD practice feel much closer to Time Surfing — without adding stress or pressure.

• It turns scattered real-world planning into a functioning project system. For a youth chess championship, I had chats, emails, volunteer names, venue logistics, catering questions, materials, and deadlines spread across different places. Mem Agent consolidated that into a live project with milestones, owners, dependencies, and parallel next actions — while keeping genuinely independent workstreams open instead of forcing everything into one artificial sequence. I could and did keep refining the system by teaching it how I wanted reviews, routines, and follow-through to work, rather than adapting my life to a fixed template.

• It gives me a working bridge between my daily Bullet Journal on my Remarkable and my digital system. Mem Agent reads those pages with varying degrees of accuracy, then cross-checks what it extracts against Mem’s own notes, tasks, projects, and dates. That makes the BuJo and the digital system partners rather than competing sources: discrepancies can become visible, useful information can be carried over, and I can keep the flexibility of writing by hand without letting the digital record quietly drift away from reality. The occasional extra work needed to decipher a page is probably more a reflection of my handwriting than of the system itself.

• It has become a personal health-data reconciliation layer rather than just a place to dump wearable numbers. I gave it a HealthLog containing journal entries, nutrition, Withings body data, and workouts, alongside and using Oura and Bevel exports; Mem Agent turns these different sources into analysis-ready tables for daily overview, sleep, cardiovascular data, activity, scores, habits, and workouts. It also preserves the actual measurement date instead of the import date and resolved an apparent Oura-versus-Bevel contradiction by recognizing that Bevel exports refer to the previous night and that Bevel’s recovery percentages are its own calculations, not Oura scores. That makes it possible to look at sleep debt, HRV, readiness, hydration, stress, and training together with what actually happened that day — without pretending the data supports medical conclusions.

• I also use Claude as a cross-checking layer for sources that Mem Agent cannot yet access directly, including Granola, Recall, and Evernote. Claude compares what those tools contain with Mem Agent’s records and surfaces information that may be missing or unclear, so I can bring it into the conversation and keep the overall system more complete.

Overall, Mem has made my work simpler, clearer, and more relaxed. It keeps important commitments and context in view, helps me think through uncertainty, and lets me use GTD in a more “Time Surfing”-way without extra pressure.

I am a big fan! I have been using Mem AI for a few months now, and I have to say that since the MCP integration, my life has changed a little bit. Everything is there now, and I’m just a question away from finding anything I need in both my professional and personal life.

I have tried many similar apps, but Mem was the perfect fit right away. btw, I barely use it for work. I mostly use it for personal stuff and it’s still worth every penny (but please don’t raise the prices!! lol)

Hi Product Hunt! I’m Emre — and I’m back.

Today I’m launching WebBrain, a free, open-source AI browser agent that lives directly inside your browser.

I built it because the most capable browser agents are generally closed source, tied to specific models, and locked behind subscriptions. WebBrain takes a different approach:

• Works with Claude, OpenAI, OpenRouter, or local models through llama.cpp

• Runs on Chrome, Firefox, and Edge

• Uses screenshots and browser accessibility data to understand and operate websites

• Keeps you in control of the model, provider, and data

Drafting this launch, I used WebBrain to write parts of it. 🙂

Ask. Act. Automate. Any LLM.

This isn’t my first attempt at making the web more programmable. In 2006, I launched Grou.ps, which grew to 350,000 communities and 12 million registered members before its assets were acquired by a publicly traded telecom company.

Later, I launched GraphJS here on Product Hunt—an open-source toolkit that let developers add social-network features to websites using a few HTML tags and JavaScript. About a year later, it was acquired by its first enterprise customer.

Now I’m bringing that same open-web philosophy to AI browser agents.

Try it: 👉 webbrain.one
Star it: ⭐ github.com/webbrain-one/webbrain

MIT licensed and open source.

Try it on this page right now — have it draft or revise your comment and actually post it. Or just ask "who built this?"

How well do local models handle more complex multi-step browser tasks compared with Claude or OpenAI? Do you see local inference becoming good enough that most everyday browser automation can eventually stay fully on-device?

I typically use @Dia Browser which has its own AI assistant built-in, but I appreciate that@emresokullu’s WebBrain extension gives people choice who use Chrome and other browsers.

The power of this approach is that the assistant has access to your sessions, cookies, and context — intact.

And you can BYOK or point it at Claude, OpenAI, OpenRouter, or even a local model through llama.cpp, then ask it to understand the page, fill forms, or take action. You can use any LLM in your browser with WebBrain.

I've known @emresokullu from back in the social web days when I hunted @Grou.ps v2 7 years ago! Nice to see him back at it!

Hi everyone!

A phone that gets more useful when you turn it face down is a pretty interesting idea.

On Pixel 11 Pro, the new HiLight system uses colored LEDs around the camera flash to tell you when a favorite contact is calling, or when @Gemini is listening, thinking, and responding, without asking you to look at the screen.

Gemini Intelligence can now handle more things in the background, from moving information across apps to taking on everyday tasks like booking a ride or placing an order.

The hardware is moving in the same direction too. Tensor G6 brings much more on-device AI compute, while the cameras, charging, and durability all got meaningful upgrades (7 years!)

Love the idea of AI making the phone a little quieter. Do more in the background, then give me just enough signal when something actually needs my attention :)

@nishant_aggarwal Not a terminal person at all, is there a way to do this without the command line, or is that the whole point?

One of the cool things about the CLI (vs MCP) is that it's a lot faster, and you get a very readable trace of the commands that the agent executed, then you can use it to build especialized document skills with greater control.

For example a very simple loop that I use when I publish a fluid doc, is to use claudecode + cli to get the questions that the users have asked on the document's QNA chat to improve the document itself.

@nishant_aggarwal When I send someone a doc, do they need an account or an app to ask it questions, or does the link just work in their browser? My clients will not sign up for anything, so this is the whole decision for me.

I'm looking forward to digging in to see how I could use it with the Kithli pitch deck as we prepare to raise a pre-seed round. But, even before that with the Revenue Leakage research project we're running with ASCs.

Hey Product Hunt, we are thrilled to launch the FluidDocs CLI. 🎉

FluidDocs are interactive documents that keep doing their job after you hit send: readers ask them questions and get answers from the document itself, and you see who opened them, how far they got, and what they asked.

The reason we built a CLI is that more than half of the 1,000+ documents created on FluidDocs, were created programmatically, by coding agents working through our MCP, not by a person clicking in the app. Our users' agents were already writing the documents. They needed a first-class way to publish and maintain them.

So now one line installs fld, and from there everything runs in the terminal:

✅ Publish a file, a zip, or a whole directory as an interactive document
✅ Turn on reader Q&A with one command
✅ Pull reader analytics back into the terminal
✅ Edit the live document later, and the same link updates
✅ --json on every command, so your coding agent can run the whole flow unattended

The best way in: install, run fld login, and publish something you already have. Or try the SpaceX S-1 as a FluidDoc at https://share.fluiddocs.ai/spacex-s1 and ask it a question a static document could never answer.

We are taking command and workflow requests all week. Tell us what your publishing flow needs and we will build it.

congrats on the launch. excited to try it out

Twenty clean runs of the same task can still hide brittleness. I would vary page state, auth/session state, latency, layout drift, and an interrupted run, then report completion, recovery, time, and cost. Reliability comes from the distribution, not the best demo.

This is actually really needed. I feel like CUA agents are way too slow (hence why I don't really use CUA/browser-use in general), so hopefully this helps us get to CUA agents that are faster :)

What task category are we missing?

What task would make you say ‘okay, agents are actually useful now’?

What’s a benchmark metric you think everyone is over-optimizing right now?

I’d love to benchmark tasks people actually do every week, not just benchmark-y tasks. What repetitive computer task would you outsource to an agent tomorrow?

Interesting question we’ve been discussing internally: should retries count? If an agent succeeds on attempt 3, is that a pass or a fail?

Would you rather see agents compete head-to-head, or see each one evaluated against a fixed passing score?

Should agents get penalized for unnecessary actions? e.g. finishing the task correctly but clicking around 40 times when 8 would do.

We’re thinking about adding more adversarial tasks. Stuff that looks easy at first but has a few traps. Any ideas?

Give us a computer task you think no current agent can reliably complete.

If it’s reproducible, we might turn the best suggestions from this thread into Coarena battles 👀

We hide model identities until after the vote because names carry a lot of baggage.

I’d be curious to know how often people’s preferred run changes once they discover which model produced it.

Would you trust a purely automated judge for computer-use tasks?

We’re using blind human preference because many real workflows don’t have a simple exact-match answer, but human judgment introduces its own inconsistencies too.

How much should recovery matter?

An agent that makes a mistake and fixes it may take longer, but that ability to recognize and recover from failure could be more valuable than a perfect run on an easy task.

What should happen when both agents complete the task, but one takes a strange or risky route?

The final output may look identical even though the trajectory tells a very different story.

Suppose Agent A finishes correctly in 20 steps and Agent B finishes correctly in 8 steps.

Should B automatically win, or should efficiency only matter when the final results are otherwise identical?

One constraint we take seriously: please don’t submit passwords, payment details, private documents, or personal data.

The best tasks are realistic without requiring anything sensitive. We want the arena to be useful, not reckless.

What business software should agents be tested on next?

We’re especially interested in tools with real multi-step workflows, not just websites where the agent finds one fact and stops.

A useful task doesn’t have to be complicated.

“Find the cancellation policy for this hotel” or “compare these two plans” can reveal a lot about whether an agent reads carefully or just clicks the first plausible result.

Would you rather test agents on short tasks with one clearly correct answer, or longer workflows where there are several acceptable ways to finish?

The longer tasks feel more realistic, but they’re also much harder to judge consistently.

I’m looking for tasks that are easy for a person but surprisingly difficult for an agent.

Anything involving ambiguous menus, changing page layouts, multiple tabs, or recovering from a wrong click is especially interesting. What would you submit?

What’s the most boring computer task you repeat every week?

Those are probably the best tests for computer-use agents, not carefully designed demos, but the annoying workflows people would genuinely like to stop doing.

A quick caveat: the leaderboard is still early, so the rankings will move as more real tasks and votes come in.

That’s intentional. We’d rather show the uncertainty and let the benchmark evolve in public than publish one polished score that looks more definitive than it really is.

Which two computer-use agents would you most want to see go head-to-head?

We’re prioritizing the next model integrations now. Specific matchups are more helpful than a general list, we want battles where the result isn’t already obvious.

A part of Coarena that may not be obvious from the landing page: every battle produces a full trajectory.

That includes the screenshots, actions, intermediate steps, and the final blind preference. We think this can become a much more useful evaluation dataset than another fixed collection of synthetic tasks.

What would researchers or agent teams want included in an export?

We’re still debating what “best agent” should mean.

Is it the one that completes the task fastest? The one that makes the fewest mistakes? Or the one that notices an error and successfully recovers?

Right now the final judgment comes from the person reviewing both runs, but I’m curious which signals you’d want to see alongside the vote.

One thing I’d love people to try: give Coarena a task you actually do at work, especially something messy that wouldn’t appear in a normal benchmark.

Two agents will attempt it, their identities stay hidden while you review the runs, and you pick the winner.

If both agents fail, tell us where. That’s honestly just as useful as a clean success.

Really like the real-tasks-over-benchmarks approach. How do you keep scoring reproducible when the environment (a site or app) changes between runs?

I've needed this for so long!

Congrats on the launch @bhaumik_lathiya @rishitmavani!

Really interesting problem as agents start taking more actions in real systems. Excited to see this future.

Curious how you test agent behavior before production today, especially scenarios that are difficult, expensive, or risky to reproduce against real infrastructure?

Hi Product Hunt — Rishit here.

A support agent reads a ticket. Buried in the message is a line the customer did not write. The agent does not read it as text, it reads it as an instruction. Eleven seconds later it has exported the customer table, read the payment vault, moved money, and closed the ticket.

Nobody broke in. The agent had every permission it used. Every log line says authorized.

I built Execlave because I have felt this gap firsthand. As a software engineer integrating agents into real systems, I kept watching the same thing happen: the pilot works, everyone is impressed, and then the agent is given access to real data and real tools — and it runs straight into security, compliance and accountability problems that nothing in the stack was built to answer.

Every tool I found watched agents after the fact. Traces, evals, dashboards, all retrospective. None of them could refuse the call. Governance only works if it sits directly in the execution path, so that is where we put it.

Execlave is a runtime gate between your agents and the systems they touch:

• Enforce — every action is evaluated against your policies before it executes. 20 policy types and four enforcement modes — block, warn, monitor, require_approval. Allow, deny, or hold for a human.

• Prove — every call, payload and verdict is kept, cryptographically signed and replayable. You reconstruct an incident from the record, not from guesswork.

• Stop — a kill switch for one agent, a team, or everything, in one click.

• Report — decisions map to SOC 2 Type II, EU AI Act, ISO 27001, GDPR, HIPAA, PCI DSS and NIST AI RMF as signed evidence, exported in one file.

Where it is honest about its limits: a gate is only as good as the policies you give it. We ship 19 types and sensible defaults, but your first week is spent deciding what your agents are actually allowed to do. We would rather say that than pretend it is magic.

There is a free plan for evaluation — one agent, 500 traces, no card — so you can point it at something you built and see the gate work before talking to anyone. It is non-commercial; production starts at $199/mo. Thanks to @fmerian for the hunt.

I am in the comments all day. If you are running agents with production access, I especially want to hear what you have had to block by hand — that is the list we build against next.

Congrats on the launch! The hard part with agents usually is that you find out about a bad action by reading logs after it already hit a real system, so moving the check to before the action makes sense to me. What happens on the agent's side when something gets blocked? Does it get enough context to retry a different way, or does the run just stop?

The audit trail is the part I'd want to poke at. When an action passes policy but the target system then fails — timeout, 503, half-written state — does that land in the trail as its own outcome, or does the record end at "authorized"? I run an MCP server over an EU government service that goes down on a regular basis, and keeping "blocked by policy", "never reached the system" and "the system said no" as three distinct outcomes turned out to matter more than the policy logic itself. An auditor asking what happened at 14:02 needs those answered differently, and most trails I've seen collapse the last two into one.

Love the concept. What was the hardest part building this?

This made me think of DORA. Having governance written down is one thing, but being able to enforce those controls while agents are actually running seems increasingly important.

Looks intereting. Congrats team!

Hey Product Hunt! 👋

Maker here. I'm co-founder of @Execlave , and I want to tell you why we actually built this.

The moment it clicked for me wasn't a big breach story. It was how ordinary the failures were.

In conversation after conversation with teams putting agents into production, the same thing kept coming up: nobody was afraid of the model going rogue. They were afraid of the boring stuff.

A permission someone forgot to revoke. A spend limit that lived in a doc but never in the code. An agent that did exactly what its access allowed, at 3am, when no one was watching.

And almost everyone had the same setup: a clear policy, and no way to enforce it on the running agent. They could tell you what the agent was supposed to do. They couldn't stop it when it didn't.

That gap, between the governance people write down and what actually holds at runtime, is what convinced us this is a company and not a feature. As agents start taking real actions inside real systems, "we have a policy" stops being enough. Something has to be able to say no at the moment it matters.

That's the problem we care about.

Two things I'd genuinely love feedback on: if you're running agents in production today, what actually stops yours from doing something it shouldn't right now? And does the "policy on paper vs enforcement at runtime" gap match what you're seeing, or not?

Would love to hear how you're thinking about it.

Simply to use, via simple prompt in Claude cli ))

Video is still the most tricky thing to make with AI so it's awesome to see this MCP making it easier!

Do you typically aim for one-shot conversions or chaining together iterative changes?

Hi everyone!

This is essentially three cameras merged into one pocket-sized body.

First, it’s a flagship 360 camera with dual 1/1.1" Sony sensors, 8K50fps, and native Dolby Vision. Second, InstaFrame 2.0 lets it work like a gimbal camera, directly outputting stabilized 4K flat video without any reframing. Third, pair it with the POV Head Tracker and the camera can follow where you look.

The new AI Director can even edit your 360 footage while the camera charges. And if that wasn’t enough, the 3D Time Capsule uses Gaussian Splatting to turn captured spaces into 3D scenes.

It’s a massive spec bump, but also a rethink of what a 360 camera can actually be used for!

During work travel in Munich had a chance to try one of Insta360 products. Was pretty cool and this one looks even better. Hopefully will have a chance to try it once :)

Wow, didnt know i needed this until I saw it. Buying you a coffee bro, this is awesome

Good job! I was missing a proper desktop client. Bonus points for making it open source. 🖤

I had this idea in 2013. Push notifications were new, endless scroll had just arrived, and what stuck with me was watching people pay real money for cosmetic items that gave them no gameplay advantage — purely to be seen at the top of something.

So the question I couldn't drop: would people spend time instead of money? Not time as a side effect of playing, but time as the actual price. Would someone sit and do nothing for fifteen minutes to be number one on a leaderboard?

That's the whole game. A black screen. A circle appears whenever it feels like it — every level roughly doubles the window, but the minimum is always half a second, so you can never settle into the wait. Tap it in time or start over at Level 1.

The part I'm proudest of is the stats screen: an interactive particle field built from your own play history, where each particle is 100ms you spent waiting. Two people at the same level get completely different images, and you can export yours.

Built solo over five months in React Native and Expo. It's free — genuinely curious how far people get, and whether anyone finds the waiting as calming as I eventually did.

Free to play, always. If you want unlimited resumes, use WORTHTHEWAIT to get a 50% discount, reducing it from $14.99 to $7.99 USD — tap Redeem Code in the Store, or https://apps.apple.com/redeem?ctx=offercodes&id=6759990913&code=WORTHTHEWAIT

I've been beating around an idea like this myself from more of a meditation angle - looking forward to trying it!

Love this concept and game! It tricks you into meditation. We all need more patience in our lives. I could see this blowing up on discord where people are steaming themselves playing this!